Digital Pre distortion Linearization Modeling Based on Cloud Platform

2018 2nd International Conference on Modeling, Simulation and Optimization Technologies and Applications (MSOTA 2018) ISBN: 978-1-60595-594-0

Digital Pre-distortion Linearization Modeling Based on Cloud Platform

Ri-na SU

, Tai-jun LIU

, Yan YE

and Gao-ming XU

1_{Faculty of Electrical Engineering and Computer Science, Ningbo University, China}

2_{College of electronic and information engineering, Ningbo University of Technology, China}

*Corresponding author

Keywords: Pre-distortion, Linearization, Parameter extraction, Cloud platform.

Abstract. To solve the problem of multiple power amplifiers digital pre-distortion linearization modeling in the concurrent processing and avoid the defects of traditional technology in processing digital pre-distortion power amplifier system, provide a method of digital pre-distortion linearization modeling based on cloud platform. Through the remote measurement of the input signal and output signal in the amplifier system and the linearization modeling on cloud computing, extract digital pre-distortion linearization model parameters, set up power amplifier system behavior model and make the amplifiers linearization.

Introduction

The main function of power amplifier (PA) is to amplify the modulated signal to the need of power; it’s an indispensable key component in modern wireless communication system. But the nonlinear characteristics of PA lead to the regeneration or extension of out-of-band signal spectrum and the interference of adjacent channel. Especially in broadband communication, it influences memory obviously and affects the normal transmission of the communication system seriously.

At present it adopts PA digital pre-distortion techniques to avoid the adjacent channel interference and signal distortion mainly. But with the development of wireless communication technology, transmission signal is from 3G to 4G, even 5G, the training data required for digital pre-distortion modeling is getting larger and larger, traditional digital pre-distortion system based on hardware chip technology will produce the loop delay problems and does not support the remote and parallel processing of parameter extraction. Also the pre-distortion algorithm based on neural network is easy to fall into local convergence; big data training cannot solve problems in the modeling process.

With the matures of cloud computing technology, provide a method of digital pre-distortion linearization modeling based on cloud platform in order to satisfy the concurrent processing requirements of the multiple power amplifier digital pre-distortion linearization and the remote measurement and control of the signal. The use of cloud computing and cloud platform can improve the efficiency of digital pre-distortion linearization. At the same time, it can also avoid purchasing a large number of expensive hardware.

The System Model

Establish the digital pre-distortion linearization system model on cloud computing for the parameter extraction of digital pre-distortion linearization. The system model includes the application server, parameter extraction server, cloud storage server, vector signal generator, spectrum analyzer, power amplifier, attenuator, coupler and load.

PSA E4448A Spectrum analyzer

power

N5769A PA Python

VSA89600

VSG E4438C

GPIO

coupler REF IN

10

MH

z

OU

T

EV

EN

T1

O

UT

TRIG IN

Signal generator

attenuator load

GP

IO

Figure 1. Digital pre-distortion system model.

First of all, VSG and PC using GPIO port to connect together, and then use the cloud platform of user interaction module software download the test signal to signal generator VSG E4438C. Baseband signal is input into the power amplifier. Power amplifier through the attenuator connects to the matched load. On the other side, the power amplifier through the coupler connects to the spectrum analyzer E4448A to show the power output. The spectrum analyzer with acquisition board and the vector signal analyzer VSA89600 can make the frequency to baseband and capture the real-time power output data.

Secondly, in order to implement cloud computing for modeling, it needs to build cloud platform with the application server, parameter extraction server and the cloud storage server. The application server deploying the computing platform of the whole system is used to deal with the complete Http requests and corresponding in the cloud; finish more power amplifier pre-distortion processing requests for centralized management, dynamic migration and task scheduling; Parameter extraction server is used to remote data transmission and finish the power amplifier nonlinear characteristic parameters extraction; the cloud storage server is used to store the nonlinear characteristic parameters of the amplifier.

Digital Pre-distortion Linearization Modeling

A digital pre-distortion linearization based on cloud platform parameter extraction method mainly contains capture the original data sequence, capture data modeling, adjust time delay, normalization of data, behavior model extraction and pre-distortion parameters extraction. Specific as follows:

(1) Use spectrometer of the acquisition card and signal generator synchronous to capture a long continuous sequence of the original input signal and power amplifier output signal, such as 20000 data points.

1 0 1 ( ) N out _out i

A A i

N   



( ( ) )( ( ) ) 0

( )

( ( ) )( ( ) ) 0

in out

N m

in out

in out

n

A A _{N m}

in out

in out

n

A n m A A n A m

C m

A n m A A n A m

       _{   }     _{   } 





Among them, A iin( ) _and Aout( )i _{each represents the input and output signal amplitude,}N _is

on behalf of the modeling data points, Ain _and Aout _{represent the average amplitude of the input}

and output signal,m is on behalf of delay adjustment mobile points, CA Ain out( )m is for the

corresponding after mutual covariance value. Find the maximum value of the CA Ain out( )m . Combine

with the actual sampling rate fs_{, signal delay time can be calculated, the computation formula is as} follows: max 1 s m f    (4)

Among them  is the delay time, mmax _{is maximum mutual covariance adjustment of} corresponding points.

(4) Normalize process the amplifier input and output data, namely the maximum power input sequence point and output sequence point adjust to 1.

(5) The normalization of data modeling can be used for the power amplifier behavior model, based on the cloud big data processing ability and the concurrent processing capability, use real value RBF neural network model. Normalization of input data is as input signals of the model, the normalized data after the power output is as model of the output signal, through the cloud of python scientific computing module, conclude the AM/AM and AM/PM nonlinear characteristic figures of model. Evaluate the model precision through the normalized mean square error (NMSE) modeling.

(6) The output data is as input signals of the model, the input data is as a model of the output signal, extract pre-distortion parameters, complete a parameter extraction.

(7) Repeat steps (1)-(6), extracted and updated parameters again, until the algorithm convergence.

The Simulation Results

-50 -40 -30 -20 -10 0 10 -55

-50 -45 -40 -35 -30 -25 -20 -15

AM/AM RBF Model

(dBm)

(dB)

Source-Gain RBF-Gain

- 5 0 - 4 0 - 3 0 - 2 0 - 1 0 0 10

50 100 150 200

AM/PM RBF Model

(dBm)

(degree)

Source-Phase RBF-Phase

Figure 2. AM/AM and AM/PM characteristic of PA.

Conclusions

Compare with the existing technology, this paper proposed the method of multiple power amplifier system concurrent produces digital pre-distortion processing request using the cloud unified centralized management, unified deployment of hardware environment of remote data acquisition system and training model. It greatly reduces the hardware cost. The digital pre-distortion use cloud computing to implement based on the analysis of the cloud data processing, storage, computing power, greatly improve the pre-distortion parameter training processing power and speed. Combine the cloud big data parallel processing technology, such as the virtualization technology, optimize technology, the better adaptive algorithm, it can meet the requirements for actual linearity model, so that the system has practical application value.

Acknowledgements

This research was financially supported by the National Natural Science Foundation of China under Grant 61571251 and 61501272, in part by the Zhejiang Provincial Public Welfare Technology Application Research Project 2016C34003, and in part by the Ningbo Natural science fund project under grant No. 2016A610110.

References

[1] Yuan X H. Behavioral modeling and digital pre-distortion of nonlinear power amplifiers with memory effects[D]. Southwest Jiaotong University, 2016.

[2] Joel H. K. Vuolevi, Timo Rahkonen and Jani P. A.Manninen,Measurement Technique for Characterizing Memory Effects in RF Power Amplifiers, IEEE Transactions on microwave theory and techniques, vol. 49, no. 8, pp. 1383-1389, Aug. 2001.

[3] J.C. Pedro and S.A. Maas, A comparative overview of microwave and wireless power-amplifier behavioral modeling approaches, IEEE Trans. Microw.Theory Tech., vol.53, no. 4, pp. 1150-1163, Apr. 2005.

[6] Junghwan, M. and K. Bumman, Enhanced Hammerstein Behavioral Model for Broadband Wireless Transmitters. IEEE Transactions on Microwave Theory and Techniques, 2011. 59(4): pp. 924-933.

[7] Mkadem, F. and S. Boumaiza, Physically Inspired Neural Network Model for RF Power Amplifier Behavioral Modeling and Digital Pre-distortion. IEEE Transactions on Microwave Theory and Techniques, 2011. 59(4): pp. 913-923.

[8] Moon, H. and R. Sedaghat, FPGA-Based adaptive digital pre-distortion for radio-over-fiber links. Microprocessors and Microsystems, 2006. 30(3): pp. 145-154.

[9] Hekkala, A., M. Lasanen, L.C. Vieira, et al. Architectures for Joint Compensation of RoF and PA with Nonideal Feedback. In Vehicular Technology Conference (VTC 2010-Spring), 2010 IEEE 71st. 2010.